Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. An apparatus for processing an image, comprising: a bilateral filter circuit configured to: determine whether a block of pixels in a raw image has directionality, the block of pixels including pixels of multiple colors; determine, responsive to determining that the block of pixels has the directionality, taps for pixels of a same color in the block by processing directional filter coefficients corresponding to the directionality and a location of each pixel within the block; and obtain each of pixel values of a filtered raw image by multiplying the taps to corresponding pixel values in the pixels of the same color and adding the multiplied values; and a demosaicing circuit configured to perform demosaicing on a version of the filtered raw image.
This invention relates to image processing, specifically improving raw image filtering and demosaicing. The problem addressed is the need for efficient directional filtering of raw image data, which contains pixels of multiple colors (e.g., Bayer pattern), followed by accurate demosaicing to produce a full-color image. The apparatus includes a bilateral filter circuit that processes raw image blocks to reduce noise while preserving edges. The circuit first analyzes a block of pixels to determine if it exhibits directionality (e.g., edges or textures with a dominant orientation). If directionality is detected, the circuit applies directional filter coefficients tailored to the block's orientation and pixel locations. These coefficients are used to compute "taps" (weighted contributions) for pixels of the same color within the block. The filtered pixel values are obtained by multiplying these taps with the original pixel values and summing the results, producing a filtered raw image with reduced noise and preserved edges. The filtered raw image is then passed to a demosaicing circuit, which reconstructs full-color pixel values from the incomplete color data in the raw image. The demosaicing process leverages the filtered image to improve color interpolation accuracy, resulting in a higher-quality final image. This approach enhances image quality by combining directional filtering with demosaicing, particularly useful in low-light or high-noise scenarios.
2. The apparatus of claim 1 , wherein the bilateral filter is further configured to determine non-directional taps by processing default filter coefficients, responsive to determining that the block of pixels has no directionality.
3. The apparatus of claim 2 , wherein the bilateral filter is further configured to determine a confidence value representing confidence in the directionality, and wherein the taps are determined by blending, using the confidence value as a weighting factor, the non-directional taps and directional taps derived from the directional filter coefficients.
4. The apparatus of claim 1 , wherein the bilateral filter circuit is further configured to detect (i) an edge in the block or (ii) locations along a direction in which a gradient change is below a threshold.
5. The apparatus of claim 4 , wherein a directional filter coefficient for each pixel is one selected from (i) an on-edge coefficient indicating that the pixel is on the edge, (ii) an on-gradient coefficient indicating that the pixel is not on the edge but along the direction in which the gradient change is below a threshold or (iii) an off-edge coefficient indicating that the pixel is not on the edge and not along the direction in which the gradient change is below the threshold.
6. The apparatus of claim 5 , wherein each of the taps is obtained by multiplying a corresponding attenuating factor to the directional filter coefficient, wherein the attenuation factor is determined by at least a corresponding curve, a location of each pixel in the block and pixel value differences.
7. The apparatus of claim 1 , wherein the bilateral filter circuit is further configured to: determine the directionality by at least determining energies each of which is derived from pixel value differences of pixels along a plurality of directions; and select an energy with a lowest value as indicating the directionality.
8. A method for processing an image, comprising: determining whether a block of pixels in a raw image has directionality, the block of pixels including pixels of multiple colors; determining, responsive to determining that the block of pixels has the directionality, taps for pixels of a same color in the block by processing directional filter coefficients corresponding to the directionality and a location of each pixel within the block; obtaining each of pixel values of a filtered raw image by multiplying the taps to corresponding pixel values in the pixels of the same color and adding the multiplied values; and demosaicing of a version of the filtered raw image.
9. The method of claim 8 , further comprising determining non-directional taps by processing default filter coefficients, responsive to determining that the block of pixels has no directionality.
10. The method of claim 9 , further comprising determining a confidence value representing confidence in the directionality, and wherein the taps are determined by blending, using the confidence value as a weighting factor, the non-directional taps and directional taps derived from the directional filter coefficients.
This invention relates to signal processing, specifically techniques for enhancing directional filtering in audio or signal processing systems. The problem addressed is improving the accuracy and adaptability of directional filtering by incorporating confidence-based blending of non-directional and directional filter coefficients. The method involves processing an input signal to determine directional filter coefficients, which are used to derive directional taps for filtering the signal. Additionally, non-directional taps are generated, typically from a standard filter or a baseline filter response. A confidence value is calculated to represent the reliability or certainty of the determined directionality. This confidence value is then used as a weighting factor to blend the non-directional taps and the directional taps. The blending process adjusts the contribution of each set of taps based on the confidence value, ensuring that the final filter output is optimized for the given signal conditions. Higher confidence in directionality results in greater reliance on directional taps, while lower confidence leads to a higher weight on non-directional taps. This adaptive approach improves filtering performance in dynamic environments where signal directionality may vary.
11. The method of claim 8 , further comprising detecting (i) an edge or (ii) locations along a direction in which a gradient change is below a threshold.
12. The method of claim 11 , further comprising selecting one from (i) an on-edge coefficient indicating that the pixel is on the edge, (ii) an on-gradient coefficient indicating that the pixel is not on the edge but along the direction in which the gradient change is below a threshold or (iii) an off-edge coefficient indicating that the pixel is not on the edge and not along the direction in which the gradient change is below the threshold.
13. The method of claim 12 , wherein each of the taps is obtained by multiplying a corresponding attenuating factor to the directional filter coefficient, wherein the attenuation factor is determined by at least a corresponding curve, a location of each pixel in the block and pixel value differences.
This invention relates to digital image processing, specifically to methods for enhancing image quality by adjusting filter coefficients in a directional filter. The problem addressed is the need to improve image sharpness and reduce artifacts, such as ringing or blurring, when applying directional filters to image blocks. Traditional directional filters often produce inconsistent results due to variations in pixel values and spatial relationships within the image block. The method involves obtaining multiple taps for a directional filter by applying an attenuating factor to each directional filter coefficient. The attenuating factor is dynamically determined based on a predefined curve, the pixel's location within the block, and differences in pixel values. This adaptive attenuation ensures that the filter coefficients are adjusted according to local image characteristics, improving edge preservation and reducing artifacts. The directional filter coefficients are first calculated, and then each coefficient is modified by multiplying it with its corresponding attenuating factor. The attenuating factor is derived from a curve that accounts for spatial and value-based variations, ensuring that the filter adapts to different regions of the image. This approach enhances image sharpness while minimizing distortions, particularly in areas with high-frequency details or abrupt transitions. The method is applicable to various image processing applications, including noise reduction, edge enhancement, and artifact suppression.
14. The method of claim 8 , wherein the directionality of the raw image is determined by determining energies each of which is derived from pixel value differences of pixels along a plurality of directions; and selecting an energy with a lowest value as indicating the directionality.
This invention relates to image processing, specifically determining the directionality of raw image data. The problem addressed is accurately identifying the dominant orientation or directionality in an image, which is useful for tasks like image alignment, feature extraction, and noise reduction. The method involves analyzing pixel value differences along multiple directions to compute energy values, where each energy value represents the variation or gradient strength in a specific direction. The direction with the lowest energy value is selected as the dominant directionality, as it indicates the least variation and thus the most consistent orientation. This approach leverages the fact that edges or structures in an image typically align along certain directions, and the direction with minimal energy corresponds to the most prominent orientation. The method may be applied in various imaging applications, including medical imaging, satellite imagery, and document scanning, where understanding image orientation is critical for further processing or analysis. The technique improves upon prior methods by providing a more robust and computationally efficient way to determine directionality without requiring extensive preprocessing or complex models.
15. An imaging system, comprising: an image capturing device configured to capture a raw image; a bilateral filter circuit configured to: determine whether a block of pixels in the raw image has directionality, the block of pixels including pixels of multiple colors; determine, responsive to determining that the block of pixels has the directionality, taps for pixels of a same color in the block by processing directional filter coefficients corresponding to the directionality and a location of each pixel within the block; and obtain each of pixel values of a filtered raw image by multiplying the taps to corresponding pixel values in the pixels of the same color and adding the multiplied values; and a demosaicing circuit configured to perform demosaicing on a version of the filtered raw image.
16. The system of claim 15 , wherein the bilateral filter is further configured to determine non-directional taps by processing default filter coefficients, responsive to determining that the block of pixels has no directionality.
17. The system of claim 16 , wherein the bilateral filter is further configured to determine a confidence value representing confidence in the directionality, and wherein the taps are determined by blending, using the confidence value as a weighting factor, the non-directional taps and directional taps derived from the directional filter coefficients.
This invention relates to image processing systems that enhance image quality by reducing noise while preserving edges and fine details. The system addresses the challenge of balancing noise reduction with detail retention, particularly in directional features like edges, which are often degraded by conventional filtering techniques. The system includes a bilateral filter that processes image data to reduce noise while maintaining sharp edges. The bilateral filter uses a combination of spatial and range domain filtering to preserve edges by applying stronger filtering in homogeneous regions and weaker filtering near edges. The filter employs directional filter coefficients to enhance directional features, such as edges, by applying stronger filtering along the edge direction while suppressing filtering perpendicular to the edge. The bilateral filter further determines a confidence value representing the certainty of the detected edge directionality. This confidence value is used to blend non-directional taps (derived from standard bilateral filtering) with directional taps (derived from directional filter coefficients). The blending is weighted by the confidence value, allowing the system to adaptively adjust the contribution of directional filtering based on the reliability of the edge direction detection. This ensures that directional enhancement is applied only when the edge direction is confidently detected, preventing artifacts in regions where directionality is uncertain. The result is an image with reduced noise and preserved fine details, particularly in directional features.
18. The system of claim 15 , wherein the bilateral filter circuit is further configured to detect (i) an edge in the block or (ii) locations along a direction in which a gradient change is below a threshold.
19. The system of claim 18 , wherein a directional filter coefficient for each pixel is one selected from (i) an on-edge coefficient indicating that the pixel is on the edge, (ii) an on-gradient coefficient indicating that the pixel is not on the edge but along the direction in which the gradient change is below the threshold or (iii) an off-edge coefficient indicating that the pixel is not on the edge and not along the direction in which the gradient change is below the threshold.
20. The system of claim 19 , wherein each of the taps is obtained by multiplying a corresponding attenuating factor to the directional filter coefficient, wherein the attenuation factor is determined by at least a corresponding curve, a location of each pixel in the block and pixel value differences.
Unknown
March 16, 2021
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.